Investigation of stratum corneum cell morphology and content using novel machine-learning image analysis.

Abstract

The morphology and content of stratum corneum (SC) cells provide information on the physiological condition of the skin. Although the morphological and biochemical properties of the SC are known, no method is available to fully access and interpret this information. This study aimed to develop a method to comprehensively decode the physiological information of the skin, based on the SC. Therefore, we established a novel image analysis technique based on artificial intelligence (AI) and multivariate analysis to predict skin conditions. SC samples were collected from participants, imaged, and annotated. Nine biomarkers were measured in the samples using enzyme-linked immunosorbent assay. The data were then used to teach machine-learning models to recognize individual SC cell regions and estimate the levels of the nine biomarkers from the images. Skin physiological indicators (e.g., skin barrier function, facial analysis, and questionnaires) were measured or obtained from the participants. Multivariate analysis, including biomarker levels ​​and structural parameters of the SC as variables, was used to estimate these physiological indicators. We established two machine-learning models. The accuracy of recognition was assessed according to the average intersection over union (0.613), precision (0.953), recall (0.640), and F-value (0.766). The predicted biomarker levels significantly correlated with the measured levels. Skin physiological indicators and questionnaire answers were predicted with strong correlations and correct answer rates. Various physiological skin conditions can be predicted from images of the SC using AI models and multivariate analysis. Our method is expected to be useful for dermatological treatment optimization.